lkml.org 
[lkml]   [2018]   [Dec]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH v4 2/2] arm64/bpf: don't allocate BPF JIT programs in module memory
On Fri, 30 Nov 2018 at 19:26, Will Deacon <will.deacon@arm.com> wrote:
>
> On Fri, Nov 23, 2018 at 11:18:04PM +0100, Ard Biesheuvel wrote:
> > The arm64 module region is a 128 MB region that is kept close to
> > the core kernel, in order to ensure that relative branches are
> > always in range. So using the same region for programs that do
> > not have this restriction is wasteful, and preferably avoided.
> >
> > Now that the core BPF JIT code permits the alloc/free routines to
> > be overridden, implement them by vmalloc()/vfree() calls from a
> > dedicated 128 MB region set aside for BPF programs. This ensures
> > that BPF programs are still in branching range of each other, which
> > is something the JIT currently depends upon (and is not guaranteed
> > when using module_alloc() on KASLR kernels like we do currently).
> > It also ensures that placement of BPF programs does not correlate
> > with the placement of the core kernel or modules, making it less
> > likely that leaking the former will reveal the latter.
> >
> > This also solves an issue under KASAN, where shadow memory is
> > needlessly allocated for all BPF programs (which don't require KASAN
> > shadow pages since they are not KASAN instrumented)
> >
> > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> > ---
> > arch/arm64/include/asm/memory.h | 5 ++++-
> > arch/arm64/net/bpf_jit_comp.c | 13 +++++++++++++
> > 2 files changed, 17 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/include/asm/memory.h b/arch/arm64/include/asm/memory.h
> > index b96442960aea..ee20fc63899c 100644
> > --- a/arch/arm64/include/asm/memory.h
> > +++ b/arch/arm64/include/asm/memory.h
> > @@ -62,8 +62,11 @@
> > #define PAGE_OFFSET (UL(0xffffffffffffffff) - \
> > (UL(1) << (VA_BITS - 1)) + 1)
> > #define KIMAGE_VADDR (MODULES_END)
> > +#define BPF_JIT_REGION_START (VA_START + KASAN_SHADOW_SIZE)
> > +#define BPF_JIT_REGION_SIZE (SZ_128M)
> > +#define BPF_JIT_REGION_END (BPF_JIT_REGION_START + BPF_JIT_REGION_SIZE)
> > #define MODULES_END (MODULES_VADDR + MODULES_VSIZE)
> > -#define MODULES_VADDR (VA_START + KASAN_SHADOW_SIZE)
> > +#define MODULES_VADDR (BPF_JIT_REGION_END)
> > #define MODULES_VSIZE (SZ_128M)
> > #define VMEMMAP_START (PAGE_OFFSET - VMEMMAP_SIZE)
> > #define PCI_IO_END (VMEMMAP_START - SZ_2M)
> > diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c
> > index a6fdaea07c63..76c2ab40c02d 100644
> > --- a/arch/arm64/net/bpf_jit_comp.c
> > +++ b/arch/arm64/net/bpf_jit_comp.c
> > @@ -940,3 +940,16 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
> > tmp : orig_prog);
> > return prog;
> > }
> > +
> > +void *bpf_jit_alloc_exec(unsigned long size)
> > +{
> > + return __vmalloc_node_range(size, PAGE_SIZE, BPF_JIT_REGION_START,
> > + BPF_JIT_REGION_END, GFP_KERNEL,
> > + PAGE_KERNEL_EXEC, 0, NUMA_NO_NODE,
> > + __builtin_return_address(0));
>
> I guess we'll want VM_IMMEDIATE_UNMAP here if Rich gets that merged.

I think akpm already queued up that patch.

> In the
> meantime, I wonder if it's worth zeroing the region in bpf_jit_free_exec()?
> (although we'd need the size information...).
>

Not sure. What exactly would that achieve?

\
 
 \ /
  Last update: 2018-11-30 20:21    [W:2.384 / U:0.612 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site